Test that will repro EndOfStreamException

ParquetReaderTest.cs

[Fact]
public void Read_plain_dictionary_data_page_that_has_no_content()
{
    using (var reader = new ParquetReader(OpenTestFile("/special/data_page_without_content.parquet"), leaveStreamOpen: false))
    {
        // if the issue has not been fixed, this line will throw EndOfStreamException
        DataColumn[] columns = reader.ReadEntireRowGroup();

        // Compare to ground truth from spark

        // # PySpark commands to get the ground truth
        // # In my repro file, the exception is thrown when processing a column called '_text'
        // # so get some details for that column
        // import pyspark.sql.functions as F
        // df = spark.read.parquet('data_page_without_content.parquet')
        // len(df.columns)
        // df.count()
        // df.withColumn('_text_len', F.length('_text')).groupby('_text_len').count().select('_text_len', 'count').collect()

        // Check dimensions are correct
        Assert.Equal(34, columns.Length);
        IEnumerable<int> rowCount = columns.Select(c => c.Data.Length).Distinct();
        Assert.Equal(new[] { 457 }, rowCount);

        // The "_text" column is where the EndOfStreamException used to be thrown
        // Check its values approximately
        DataColumn textCol = columns.Single(c => c.Field.Name == "_text");
        var histogram = from s in textCol.Data.Cast<string>()
                    group s by s?.Length into grp
                    select new { stringLength = grp.Key, count = grp.Count() };
        Assert.Equal(4, histogram.Count());
        Assert.Contains(new { stringLength = (int?)null, count = 387 }, histogram);
        Assert.Contains(new { stringLength = (int?)3877, count = 1 }, histogram);
        Assert.Contains(new { stringLength = (int?)9245, count = 1 }, histogram);
        Assert.Contains(new { stringLength = (int?)17476, count = 68 }, histogram);
    }
}

This gist belongs with this PR: aloneguid/parquet-dotnet#95

Quest staff:
The data_page_without_content.parquet is located in the MDS repo (as 'sqltempdbcontention_has_data_page_without_content.parquet'). It contains some implementation SQL of DS so we are not sharing it. See SCMM-3313